The conversion of woody biomass is studied by means of a layer-based model for thermally-thick biomass particles (Thunman et al. 2002, Ström et al. 2013). The model implementation is successfully validated against experiments that study particle conversion in a drop tube reactor. After this validation step, this work focuses on the well-known problem of grid dependence of two-phase numerical simulations using the standard Euler–Lagrange (EL) framework. This issue is addressed and quantified by comparing EL data that models the particle boundary layers to corresponding simulations which fully resolve these boundary layers (fully-resolved, FR, simulations). A comparison methodology for the conceptually different FR and EL approaches by extracting the heat transfer coefficient from the detailed FR simulations is proposed and confirms that the EL results are strongly grid-dependent. This issue is overcome by applying a set of coarse-graining methods for the EL framework. Two coarse-graining methods are evaluated, a previously suggested diffusion-based method (DBM) and a new approach based on moving averages referred to as MAM. It is shown that both DBM and MAM can successfully recover the detailed FR data for pure particle heating for a case where the grid size is half the particle diameter, i.e. when the standard EL method fails. Both coarse-graining methods also give improved results for an EL simulation that considers the more complex combined physics of particle heating, drying and devolatilisation, given that the CG model parameters that scale the corresponding CG interaction volumes are sufficiently large. Based on the available FR data, recommended model parameter ranges for DBM and MAM are provided as a function of normalised boundary layer thickness. The novel MAM approach is shown to be significantly more efficient than the DBM and therefore suitable for future EL simulations with multiple particles.